Re:The road to realistic full-body deepfakes
by gcanyon on Thursday September 22, 2022 @05:58PM
(#32945163)
Once you accept lossy compression, it becomes a question of what level and type of "lossy" you're willing to accept, and how clever the "compression" algorithm can be.
If I want to compress the movie Thunderball -- a sufficiently clever "compression" algorithm could start with the synopsis at https://en.wikipedia.org/wiki/Thunderball_(film) add in some images of Sean Connery, and generate the film. That's...maybe a 100K to 1 compression ratio?
If the algorithm itself understands "Sean Connery" then you could (theoretically) literally feed in the text description and achieve a reasonable result. I've seen Thunderball, but it was years ago and I don't remember the plot (boats?). I'd know the result was different, but I likely wouldn't be able to point to anything specific.
If I want to compress the movie Thunderball -- a sufficiently clever "compression" algorithm could start with the synopsis at https://en.wikipedia.org/wiki/Thunderball_(film) add in some images of Sean Connery, and generate the film. That's...maybe a 100K to 1 compression ratio?
If the algorithm itself understands "Sean Connery" then you could (theoretically) literally feed in the text description and achieve a reasonable result. I've seen Thunderball, but it was years ago and I don't remember the plot (boats?). I'd know the result was different, but I likely wouldn't be able to point to anything specific.